top of page
Glossy 3D 48PEGASUS Logo.png

An AI Order-Status Assistant That Explains What Is Happening

Oct 27
4 min read

Updated: Nov 12

“Where is my order?” sounds like a simple question. The answer becomes less simple when one item is still being made, another has shipped, and a courier link has not updated since yesterday.


An AI order-status assistant can turn scattered records into a clear explanation. Its job is to read authorised order and delivery information, explain what those records mean, and offer the next appropriate step. It cannot see inside a delivery van or guarantee an arrival time that the carrier has not confirmed.


Cardboard cartons stacked inside an industrial warehouse
Stock photograph for illustration. Photo: Freek Wolsink / Pexels.

For retailers, manufacturers, and businesses selling made-to-order products, this can be a useful first automation project because the question is common and the source data is usually identifiable.


Start with the order record


The assistant needs a trusted record for each stage: payment received, preparation underway, dispatched, handed to the carrier, or delivered. These stages should come from the commerce or fulfilment system, rather than a model's interpretation of an old email.


Some orders have several shipments. Shopify, for example, documents a fulfilment object containing shipped items, quantities, carrier details, tracking numbers, and tracking links. It also explicitly allows multiple fulfilments for one order. A well-designed assistant must preserve that distinction. Shopify fulfilment documentation


AI provides the explanation. A normal database lookup provides the facts. That separation is what keeps a friendly answer grounded.


An illustrative customer conversation


Consider a fictional homeware shop. A verified customer asks, “My lamp arrived, but where is the shade?” The order contains two items with separate dispatch records.


A useful answer could be:


“Your lamp and shade are shipping separately. The lamp is marked delivered today. The shade is packed, but the carrier has not recorded collection yet. Its current dispatch estimate is Thursday. I checked the order record at 10:15. Would you like the tracking link for the shade?”


The answer should show a genuine link when one exists. If no label has been created, it should say so. “Packed” must not become “on its way,” and a dispatch estimate must not become a delivery promise.


An internal exception record might contain the order reference, affected item, last confirmed event, event time, and the team's next required action. Staff can then investigate without asking the customer to start again.


The inputs and connections required


A workable first version needs:


  • A verified way to identify the customer and their order

  • Read access to order lines and fulfilment records

  • Carrier tracking information where the business has permission to use it

  • Clear definitions for internal order stages

  • Approved delivery, collection, cancellation, and returns guidance

  • A staffed route for exceptions


An application programming interface, or API, is a controlled connection through which one system requests information from another. It is usually preferable to copying order data manually into a chatbot. The integration should ask for the narrowest access it needs.


Order records contain personal information. Shopify's own protected-customer-data requirements emphasise minimisation, transparency, and security, with additional access requirements for certain customer fields. Those are platform-specific requirements, not a blanket compliance guarantee. Shopify protected customer data guidance


How the workflow should operate


  1. Verify access. Use an authenticated account or an approved verification process. An order number alone should not expose someone else's delivery details.

  2. Retrieve the current order. Confirm the relevant items and whether there are separate shipments.

  3. Retrieve the latest available delivery events. Keep the carrier event time separate from the time the assistant fetched it.

  4. Apply business rules. Determine whether the order is within its expected window or needs an exception queue.

  5. Write a short answer. State the verified stage, useful timing information, uncertainty, and next step.

  6. Escalate when necessary. Attach the records to a case that a person can actually own.


Updates can arrive through webhooks, which are automatic notifications one system sends when something changes. Even then, integrations can miss or delay events. Before answering a fresh customer question, a direct check of the current record may be appropriate.


Handle awkward cases honestly


A carrier marked a parcel delivered, but the customer says it is missing. The assistant should record both facts and follow the business's approved investigation process. It should not accuse the customer of being mistaken or promise a replacement without authority.


If the carrier site is unavailable, say that tracking cannot currently be checked. Give the last confirmed event and its timestamp if it is still useful. An honest partial answer is preferable to an invented update.


Address changes, cancellations after dispatch, refunds, and replacement orders need separate permission and transaction rules. A read-only status assistant should not acquire those powers simply because a customer asks in the same conversation.


Keep personal details out of public chat and notifications. A status message rarely needs to repeat the customer's full address. Test whether someone changing a reference number can view another order, and ensure logs do not become an unprotected copy of the order database.


Pilot one fulfilment path


Start with one store, one delivery method, and domestic orders with reliable tracking. Include single-item orders, split shipments, delayed dispatches, missing carrier events, and delivered-but-not-received cases in testing.


Measure answer accuracy against the underlying records, repeat contacts about the same order, and time spent on straightforward status questions. Count incorrect delivery promises as a separate failure, even if the customer initially rated the answer positively.


Also monitor how often the assistant cannot retrieve a current record. A polished conversation can conceal an unreliable integration unless that failure has its own measure.


For a first discussion with 48PEGASUS, bring the order stages your team uses and a few anonymised examples of difficult delivery questions. The most valuable design work is agreeing what each stage really means, what the customer should hear, and where a person takes over.



Comments


bottom of page